But your claim that “any OS will start to show trouble” is not how it’s supposed to work, nor how it used to work before 24/7 connections, nor even how it should work assuming you’re ok with phone-home daemons.
An OS in general has a few layers with caches and monitors and resolvers etc, and if you block a few of them, in a partial manner, they tend to get in to an extreme version of the bus bunching problem. Windows still does that with their 'network identification' where sometimes something goes wrong in the probing process and the connection hangs on "identifying" indefinitely. And that's just a silly "optional" service (which should default to the public profile and only change from there instead of defaulting to no connection at all).
Seems to be an interesting problem: if it works fine with no connection and fine with a complete connection but not 'in between' (which is the best way I can describe it so far) you'd think it must be some common library or component in a network stack that causes this. macOS has some reachability system that might be in play here, perhaps if it flags the network as 'reachable' but then gets REJECT'ed it goes bad? Or the other way around: marks network as 'unreachable' but traffic flows anyway?